Exploring phonetic category structure with Markov chain Monte Carlo
Loading...
Date
2008-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Research in cognitive psychology on how humans mentally represent phonetic categories has demonstrated that these categories have a graded internal structure, meaning that some exemplars of the category are considered subjectively better, or more representative, members of the category than are others. Since many phenomena in speech perception are sensitive to this graded category structure, the determination of this structure is a goal of research in the area. The goal of this thesis is to apply a new experimental methodology to explore the mental representations of a phonetic category. The procedure exploits a connection between human choice behavior and Markov chain Monte Carlo (MCMC) algorithms to sample from an individual's mental representation of a phonetic category, taken here to be a probability distribution over speech sounds. Stimuli for the project consisted of 81 computer synthesized examples of the /i/ phonetic category. A computer program repeatedly presented pairs of these stimuli to subjects in accordance with the requirements of MCMC algorithms, and subjects’ choices mimicked an MCMC acceptance function. The exemplars of /i/ chosen by subjects during the course of the experiment were used to estimate their mental representations of the /i/ phonetic category. The phonetic category structures estimated by the psychological MCMC procedure show that subjects tended to choose stimuli at the edge of the acoustic stimulus space as the best examples of the /i/ phonetic category. This result has not always been obtained with similar stimulus sets and a goodness rating technique, and some implications for future use of the procedure are discussed.
Description
Keywords
cognitive psychology, phonetic category structure, mcmc, speech perception